This is for people running Sanity Agent Context MCP who need to tune what their agent knows beyond the auto-generated schema. It walks you through an interactive session to build the Instructions field, focusing on pure deltas like misleading field names, reference chains the schema doesn't make obvious, or data quality issues. The workflow enforces a critical discipline: never generalize from small samples, verify every claim with queries, and keep existing instructions if they're good. You can work with write access to iterate on a draft context doc, or test everything via URL params if you only have read access. The skill's opinionated about one thing: if the schema already says it, don't repeat it in instructions.
npx skills add https://github.com/sanity-io/agent-context --skill dial-your-context